Discover more from Kamil’s newsletter
Meep #1. Full-sentence autocomplete on your computer using AI.
Let's build a product in public!
Do you ever get this feeling? You just had an idea. A special idea! An idea - which as soon as you turn it into reality - will make the world a much better place.
I just had this feeling myself. I also had it many times before, and what usually followed was frustration and disappointment. I don’t like being frustrated and disappointed.
Let’s be smarter this time. Let’s figure out if the idea is actually any good before jumping into implementation and wasting the next 6 months of life.
So, what’s the idea?
Say hi to Meep. Who’s Meep? Meep is the codename - subject to change - for the product I’m envisioning.
Here’s a gif of Meep in action that I made 5 years ago (yes, it’s really been 5 years - and no, I didn’t build the product back then, this is only a simulation):
Which made the 2018 version of me wonder - why is this possible only in these few apps, and not in every app or website out there? Available e.g. via a Chrome extension, like Grammarly.
After some time, the 2018 me figured out the answer: because AI language models are inaccurate and expensive and slow and you need an army of genius nerds (sending 💕 to AI/ML researchers and engineers) to get it all to work.
Why revisit now?
Because in 2023 we’re on a different planet when it comes to AI. Now, even a halfwit such as myself can grab OpenAI’s API, do some prompt engineering and get results that are highly accurate, fast and affordable. My main 2018 implementation blocker is gone.
Does it mean that it’s all trivial now and we can start packing our shit en route to the moon? No. The concept is promising, but this is where research should start rather than end.
Let’s recap where I am
I have an early concept: A tool that uses AI to elegantly provide sentence completion suggestions. Works on any website, delivered as a Chrome extension.
Here is what I want to find out next:
Who is the ideal customer? I know that similarly to Grammarly basically anyone that types on a computer could benefit, but how should I think about narrowing down who I reach out to 1) for whom the product makes the biggest impact and 2) who is willing to pay quickly, without a massive sales effort?
How can I reach the prospective customer fast? Could this method also work at scale?
Can this really make a big impact? The suggestions won’t be correct 100% of the time, but on balance, they should be clearly more useful than distracting.
Making a Chrome extension is easy, but what does it take to hook into all the input fields on each website? The end product should support input fields, text areas, content editable divs and God knows what else. Would any of the websites mind Meep hacking into their input fields?
What’s the right way to think about pricing? OpenAI isn’t free, my time also isn’t.
How to name the product? There are a couple tools named “Meep” - none of them are really competitive, but maybe there’s a better name?
How should I think about data privacy? Meep has to grab and process what the user is typing to produce a useful completion. OpenAI has their data policy. What policy do I need on Meep’s layer? Ideally I’d minimize data retention, but I also need some way to monitor if the product is working, and if the suggested completions are reasonable and useful.
What’s an efficient way to get the answers to all these questions?
What am I missing?
The most important thing now is to establish contact with potential customers, which means working on questions 1, 2 and 3. I’ll need:
A simple and clear way that communicates the concept.
To share the concept in the right places so that it can be discovered.
A way for people to gather and contact those who are interested (some sort of waitlist).
In post #2 - coming hopefully soon - I’ll report on my progress and new thinking. Onwards!
Thanks for reading my newsletter! Subscribe for free to receive new posts and support my work.